This message was deleted.

Committing the volume head only into the backing image is not enough since the volume data consists of the volume head plus or snapshot files. I would not recommended to manually coalesce all snapshot files plus the head into the backing image. It may be complicated. What I am thinking is improving https://longhorn.io/docs/1.7.2/advanced-resources/data-recovery/export-from-replica/ so that it can fit backing image usage cc @icy-agency-38675

Hi @famous-shampoo-18483. Unfortunately the Harvester cluster appears to be completely broken, it failed in the middle of an upgrade... long story. I think I can see from the metadata files in the PVC folders that they describe how the snapshot layers and volume head files fit together overlaying the backing file. I haven't discovered how to commit each layer onto the backing layer in turn. If there was a command, similar to qemu-img, that commit layers then I could recover some data. Otherwise, I'm happy to write some code if I know how the layering works.

I am planning to modify this script to fit the backing image case, then build a new image, and run the command mentioned in https://longhorn.io/docs/1.7.2/advanced-resources/data-recovery/export-from-replica/. This does not rely on k8s cluster while recovering a replica directory to a block device.

Thank you.